@Malaik
As @Yosi_Neuman mentions, the definition of availability is where it should start, and as almost always is the case, it depends 😁
If you are trying to factor in bad things that might happen before your servers, than synthetic is almost the only way to go. @Babar_Qayyum mentions though "Unexpected low traffic", as it also helps to track these type of issues, but you will be blind on the causes.
If there are no defined ways to measure availability, you can start with some simple ones. These are some of the ones we use with our clients, and that @Yosi_Neuman has not mentioned already:
- Number of HTTP/500 errors. They almost always indicate a severe problem. You will probably notice some coming from attacks if site is public...
- Define good "key user actions" and "key requests" and eventually define special SLOs that should be obtained.
- Define conversion goals and what is expected.
Now, when these are asked by Management/BusinessAreas, they normally just ask for availability. You have than the chance to explain that availability is a much broader topic today... And in can be a never ending discussion, as you have to define if an availability event occurs when:
- The whole site goes down?
- Some server instance is having problems and others ain't?
- Some particular user is having problems?
- If bots are having problems?
- If it is too slow (what is too slow?)?
- If everything is working, but your site suddenly is not making money?
- (and the list goes on and on)
Almost all of these issues can be tracked easily by Dynatrace, but you have to give Management a perspective that nowadays measuring availability is an engineering task, and not a particularly easy one. Take synthetics, for instance: how frequently will you measure? One hour? One minute? One second? In the latter case, what is availability if that synthetic measurement worked, and at the same millisecond, another user measurement failed?
Antonio Sousa
